Fail to run the multi-gpu case in the master branch

[richardxs]

I have tried the following command torchrun --nproc_per_node=4 train.py --dataroot ./datasets/facades --name facades_pix2pix --model pix2pix --netG unet_256 --direction BtoA --lambda_L1 100 --dataset_mode aligned --pool_size 0, but it does not work.